4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #include <sys/dsl_pool.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_prop.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dnode.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
37 #include <sys/zfs_context.h>
38 #include <sys/fs/zfs.h>
39 #include <sys/zfs_znode.h>
40 #include <sys/spa_impl.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/zil_impl.h>
44 typedef int (scrub_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_t *);
46 static scrub_cb_t dsl_pool_scrub_clean_cb;
47 static dsl_syncfunc_t dsl_pool_scrub_cancel_sync;
49 int zfs_scrub_min_time = 1; /* scrub for at least 1 sec each txg */
50 int zfs_resilver_min_time = 3; /* resilver for at least 3 sec each txg */
51 boolean_t zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
53 extern int zfs_txg_timeout;
55 static scrub_cb_t *scrub_funcs[SCRUB_FUNC_NUMFUNCS] = {
57 dsl_pool_scrub_clean_cb
60 #define SET_BOOKMARK(zb, objset, object, level, blkid) \
62 (zb)->zb_objset = objset; \
63 (zb)->zb_object = object; \
64 (zb)->zb_level = level; \
65 (zb)->zb_blkid = blkid; \
70 dsl_pool_scrub_setup_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
72 dsl_pool_t *dp = arg1;
73 enum scrub_func *funcp = arg2;
74 dmu_object_type_t ot = 0;
75 boolean_t complete = B_FALSE;
77 dsl_pool_scrub_cancel_sync(dp, &complete, cr, tx);
79 ASSERT(dp->dp_scrub_func == SCRUB_FUNC_NONE);
80 ASSERT(*funcp > SCRUB_FUNC_NONE);
81 ASSERT(*funcp < SCRUB_FUNC_NUMFUNCS);
83 dp->dp_scrub_min_txg = 0;
84 dp->dp_scrub_max_txg = tx->tx_txg;
86 if (*funcp == SCRUB_FUNC_CLEAN) {
87 vdev_t *rvd = dp->dp_spa->spa_root_vdev;
89 /* rewrite all disk labels */
90 vdev_config_dirty(rvd);
92 if (vdev_resilver_needed(rvd,
93 &dp->dp_scrub_min_txg, &dp->dp_scrub_max_txg)) {
94 spa_event_notify(dp->dp_spa, NULL,
95 ESC_ZFS_RESILVER_START);
96 dp->dp_scrub_max_txg = MIN(dp->dp_scrub_max_txg,
100 /* zero out the scrub stats in all vdev_stat_t's */
101 vdev_scrub_stat_update(rvd,
102 dp->dp_scrub_min_txg ? POOL_SCRUB_RESILVER :
103 POOL_SCRUB_EVERYTHING, B_FALSE);
105 dp->dp_spa->spa_scrub_started = B_TRUE;
108 /* back to the generic stuff */
110 if (dp->dp_blkstats == NULL) {
112 kmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
114 bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
116 if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB)
117 ot = DMU_OT_ZAP_OTHER;
119 dp->dp_scrub_func = *funcp;
120 dp->dp_scrub_queue_obj = zap_create(dp->dp_meta_objset,
121 ot ? ot : DMU_OT_SCRUB_QUEUE, DMU_OT_NONE, 0, tx);
122 bzero(&dp->dp_scrub_bookmark, sizeof (zbookmark_t));
123 dp->dp_scrub_restart = B_FALSE;
124 dp->dp_spa->spa_scrub_errors = 0;
126 VERIFY(0 == zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
127 DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
128 &dp->dp_scrub_func, tx));
129 VERIFY(0 == zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
130 DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
131 &dp->dp_scrub_queue_obj, tx));
132 VERIFY(0 == zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
133 DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
134 &dp->dp_scrub_min_txg, tx));
135 VERIFY(0 == zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
136 DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
137 &dp->dp_scrub_max_txg, tx));
138 VERIFY(0 == zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
139 DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
140 &dp->dp_scrub_bookmark, tx));
141 VERIFY(0 == zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
142 DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
143 &dp->dp_spa->spa_scrub_errors, tx));
145 spa_history_internal_log(LOG_POOL_SCRUB, dp->dp_spa, tx, cr,
146 "func=%u mintxg=%llu maxtxg=%llu",
147 *funcp, dp->dp_scrub_min_txg, dp->dp_scrub_max_txg);
151 dsl_pool_scrub_setup(dsl_pool_t *dp, enum scrub_func func)
153 return (dsl_sync_task_do(dp, NULL,
154 dsl_pool_scrub_setup_sync, dp, &func, 0));
159 dsl_pool_scrub_cancel_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
161 dsl_pool_t *dp = arg1;
162 boolean_t *completep = arg2;
164 if (dp->dp_scrub_func == SCRUB_FUNC_NONE)
167 mutex_enter(&dp->dp_scrub_cancel_lock);
169 if (dp->dp_scrub_restart) {
170 dp->dp_scrub_restart = B_FALSE;
171 *completep = B_FALSE;
174 /* XXX this is scrub-clean specific */
175 mutex_enter(&dp->dp_spa->spa_scrub_lock);
176 while (dp->dp_spa->spa_scrub_inflight > 0) {
177 cv_wait(&dp->dp_spa->spa_scrub_io_cv,
178 &dp->dp_spa->spa_scrub_lock);
180 mutex_exit(&dp->dp_spa->spa_scrub_lock);
181 dp->dp_spa->spa_scrub_started = B_FALSE;
182 dp->dp_spa->spa_scrub_active = B_FALSE;
184 dp->dp_scrub_func = SCRUB_FUNC_NONE;
185 VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
186 dp->dp_scrub_queue_obj, tx));
187 dp->dp_scrub_queue_obj = 0;
188 bzero(&dp->dp_scrub_bookmark, sizeof (zbookmark_t));
190 VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
191 DMU_POOL_SCRUB_QUEUE, tx));
192 VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
193 DMU_POOL_SCRUB_MIN_TXG, tx));
194 VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
195 DMU_POOL_SCRUB_MAX_TXG, tx));
196 VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
197 DMU_POOL_SCRUB_BOOKMARK, tx));
198 VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
199 DMU_POOL_SCRUB_FUNC, tx));
200 VERIFY(0 == zap_remove(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
201 DMU_POOL_SCRUB_ERRORS, tx));
203 spa_history_internal_log(LOG_POOL_SCRUB_DONE, dp->dp_spa, tx, cr,
204 "complete=%u", *completep);
206 /* below is scrub-clean specific */
207 vdev_scrub_stat_update(dp->dp_spa->spa_root_vdev, POOL_SCRUB_NONE,
210 * If the scrub/resilver completed, update all DTLs to reflect this.
211 * Whether it succeeded or not, vacate all temporary scrub DTLs.
213 vdev_dtl_reassess(dp->dp_spa->spa_root_vdev, tx->tx_txg,
214 *completep ? dp->dp_scrub_max_txg : 0, B_TRUE);
215 if (dp->dp_scrub_min_txg && *completep)
216 spa_event_notify(dp->dp_spa, NULL, ESC_ZFS_RESILVER_FINISH);
217 spa_errlog_rotate(dp->dp_spa);
220 * We may have finished replacing a device.
221 * Let the async thread assess this and handle the detach.
223 spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER_DONE);
225 dp->dp_scrub_min_txg = dp->dp_scrub_max_txg = 0;
226 mutex_exit(&dp->dp_scrub_cancel_lock);
230 dsl_pool_scrub_cancel(dsl_pool_t *dp)
232 boolean_t complete = B_FALSE;
234 return (dsl_sync_task_do(dp, NULL,
235 dsl_pool_scrub_cancel_sync, dp, &complete, 3));
239 dsl_free(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp,
240 zio_done_func_t *done, void *private, uint32_t arc_flags)
243 * This function will be used by bp-rewrite wad to intercept frees.
245 return (arc_free(pio, dp->dp_spa, txg, (blkptr_t *)bpp,
246 done, private, arc_flags));
250 bookmark_is_zero(const zbookmark_t *zb)
252 return (zb->zb_objset == 0 && zb->zb_object == 0 &&
253 zb->zb_level == 0 && zb->zb_blkid == 0);
256 /* dnp is the dnode for zb1->zb_object */
258 bookmark_is_before(dnode_phys_t *dnp, const zbookmark_t *zb1,
259 const zbookmark_t *zb2)
261 uint64_t zb1nextL0, zb2thisobj;
263 ASSERT(zb1->zb_objset == zb2->zb_objset);
264 ASSERT(zb1->zb_object != -1ULL);
265 ASSERT(zb2->zb_level == 0);
268 * A bookmark in the deadlist is considered to be after
271 if (zb2->zb_object == -1ULL)
274 /* The objset_phys_t isn't before anything. */
278 zb1nextL0 = (zb1->zb_blkid + 1) <<
279 ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT));
281 zb2thisobj = zb2->zb_object ? zb2->zb_object :
282 zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT);
284 if (zb1->zb_object == 0) {
285 uint64_t nextobj = zb1nextL0 *
286 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT;
287 return (nextobj <= zb2thisobj);
290 if (zb1->zb_object < zb2thisobj)
292 if (zb1->zb_object > zb2thisobj)
294 if (zb2->zb_object == 0)
296 return (zb1nextL0 <= zb2->zb_blkid);
300 scrub_pause(dsl_pool_t *dp, const zbookmark_t *zb)
305 if (dp->dp_scrub_pausing)
306 return (B_TRUE); /* we're already pausing */
308 if (!bookmark_is_zero(&dp->dp_scrub_bookmark))
309 return (B_FALSE); /* we're resuming */
311 /* We only know how to resume from level-0 blocks. */
312 if (zb->zb_level != 0)
315 mintime = dp->dp_scrub_isresilver ? zfs_resilver_min_time :
317 elapsed_ticks = lbolt64 - dp->dp_scrub_start_time;
318 if (elapsed_ticks > hz * zfs_txg_timeout ||
319 (elapsed_ticks > hz * mintime && txg_sync_waiting(dp))) {
320 dprintf("pausing at %llx/%llx/%llx/%llx\n",
321 (longlong_t)zb->zb_objset, (longlong_t)zb->zb_object,
322 (longlong_t)zb->zb_level, (longlong_t)zb->zb_blkid);
323 dp->dp_scrub_pausing = B_TRUE;
324 dp->dp_scrub_bookmark = *zb;
330 typedef struct zil_traverse_arg {
332 zil_header_t *zta_zh;
333 } zil_traverse_arg_t;
337 traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
339 zil_traverse_arg_t *zta = arg;
340 dsl_pool_t *dp = zta->zta_dp;
341 zil_header_t *zh = zta->zta_zh;
344 if (bp->blk_birth <= dp->dp_scrub_min_txg)
347 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa))
350 zb.zb_objset = zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET];
353 zb.zb_blkid = bp->blk_cksum.zc_word[ZIL_ZC_SEQ];
354 VERIFY(0 == scrub_funcs[dp->dp_scrub_func](dp, bp, &zb));
359 traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
361 if (lrc->lrc_txtype == TX_WRITE) {
362 zil_traverse_arg_t *zta = arg;
363 dsl_pool_t *dp = zta->zta_dp;
364 zil_header_t *zh = zta->zta_zh;
365 lr_write_t *lr = (lr_write_t *)lrc;
366 blkptr_t *bp = &lr->lr_blkptr;
369 if (bp->blk_birth <= dp->dp_scrub_min_txg)
372 if (claim_txg == 0 || bp->blk_birth < claim_txg)
375 zb.zb_objset = zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET];
376 zb.zb_object = lr->lr_foid;
377 zb.zb_level = BP_GET_LEVEL(bp);
378 zb.zb_blkid = lr->lr_offset / BP_GET_LSIZE(bp);
379 VERIFY(0 == scrub_funcs[dp->dp_scrub_func](dp, bp, &zb));
384 traverse_zil(dsl_pool_t *dp, zil_header_t *zh)
386 uint64_t claim_txg = zh->zh_claim_txg;
387 zil_traverse_arg_t zta = { dp, zh };
391 * We only want to visit blocks that have been claimed but not yet
392 * replayed (or, in read-only mode, blocks that *would* be claimed).
394 if (claim_txg == 0 && spa_writeable(dp->dp_spa))
397 zilog = zil_alloc(dp->dp_meta_objset, zh);
399 (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, &zta,
406 scrub_visitbp(dsl_pool_t *dp, dnode_phys_t *dnp,
407 arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
410 arc_buf_t *buf = NULL;
412 if (bp->blk_birth <= dp->dp_scrub_min_txg)
415 if (scrub_pause(dp, zb))
418 if (!bookmark_is_zero(&dp->dp_scrub_bookmark)) {
420 * If we already visited this bp & everything below (in
421 * a prior txg), don't bother doing it again.
423 if (bookmark_is_before(dnp, zb, &dp->dp_scrub_bookmark))
427 * If we found the block we're trying to resume from, or
428 * we went past it to a different object, zero it out to
429 * indicate that it's OK to start checking for pausing
432 if (bcmp(zb, &dp->dp_scrub_bookmark, sizeof (*zb)) == 0 ||
433 zb->zb_object > dp->dp_scrub_bookmark.zb_object) {
434 dprintf("resuming at %llx/%llx/%llx/%llx\n",
435 (longlong_t)zb->zb_objset,
436 (longlong_t)zb->zb_object,
437 (longlong_t)zb->zb_level,
438 (longlong_t)zb->zb_blkid);
439 bzero(&dp->dp_scrub_bookmark, sizeof (*zb));
443 if (BP_GET_LEVEL(bp) > 0) {
444 uint32_t flags = ARC_WAIT;
447 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
449 err = arc_read(NULL, dp->dp_spa, bp, pbuf,
450 arc_getbuf_func, &buf,
451 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
453 mutex_enter(&dp->dp_spa->spa_scrub_lock);
454 dp->dp_spa->spa_scrub_errors++;
455 mutex_exit(&dp->dp_spa->spa_scrub_lock);
460 for (i = 0; i < epb; i++, cbp++) {
463 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
465 zb->zb_blkid * epb + i);
466 scrub_visitbp(dp, dnp, buf, cbp, &czb);
468 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
469 uint32_t flags = ARC_WAIT;
470 dnode_phys_t *child_dnp;
472 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
474 err = arc_read(NULL, dp->dp_spa, bp, pbuf,
475 arc_getbuf_func, &buf,
476 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
478 mutex_enter(&dp->dp_spa->spa_scrub_lock);
479 dp->dp_spa->spa_scrub_errors++;
480 mutex_exit(&dp->dp_spa->spa_scrub_lock);
483 child_dnp = buf->b_data;
485 for (i = 0; i < epb; i++, child_dnp++) {
486 for (j = 0; j < child_dnp->dn_nblkptr; j++) {
489 SET_BOOKMARK(&czb, zb->zb_objset,
490 zb->zb_blkid * epb + i,
491 child_dnp->dn_nlevels - 1, j);
492 scrub_visitbp(dp, child_dnp, buf,
493 &child_dnp->dn_blkptr[j], &czb);
496 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
497 uint32_t flags = ARC_WAIT;
501 err = arc_read_nolock(NULL, dp->dp_spa, bp,
502 arc_getbuf_func, &buf,
503 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
505 mutex_enter(&dp->dp_spa->spa_scrub_lock);
506 dp->dp_spa->spa_scrub_errors++;
507 mutex_exit(&dp->dp_spa->spa_scrub_lock);
513 traverse_zil(dp, &osp->os_zil_header);
515 for (j = 0; j < osp->os_meta_dnode.dn_nblkptr; j++) {
518 SET_BOOKMARK(&czb, zb->zb_objset, 0,
519 osp->os_meta_dnode.dn_nlevels - 1, j);
520 scrub_visitbp(dp, &osp->os_meta_dnode, buf,
521 &osp->os_meta_dnode.dn_blkptr[j], &czb);
525 (void) scrub_funcs[dp->dp_scrub_func](dp, bp, zb);
527 (void) arc_buf_remove_ref(buf, &buf);
531 scrub_visit_rootbp(dsl_pool_t *dp, dsl_dataset_t *ds, blkptr_t *bp)
535 SET_BOOKMARK(&zb, ds ? ds->ds_object : 0, 0, -1, 0);
536 scrub_visitbp(dp, NULL, NULL, bp, &zb);
540 dsl_pool_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
542 dsl_pool_t *dp = ds->ds_dir->dd_pool;
544 if (dp->dp_scrub_func == SCRUB_FUNC_NONE)
547 if (dp->dp_scrub_bookmark.zb_objset == ds->ds_object) {
548 SET_BOOKMARK(&dp->dp_scrub_bookmark, -1, 0, 0, 0);
549 } else if (zap_remove_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
550 ds->ds_object, tx) != 0) {
554 if (ds->ds_phys->ds_next_snap_obj != 0) {
555 VERIFY(zap_add_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
556 ds->ds_phys->ds_next_snap_obj, tx) == 0);
558 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
562 dsl_pool_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx)
564 dsl_pool_t *dp = ds->ds_dir->dd_pool;
566 if (dp->dp_scrub_func == SCRUB_FUNC_NONE)
569 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0);
571 if (dp->dp_scrub_bookmark.zb_objset == ds->ds_object) {
572 dp->dp_scrub_bookmark.zb_objset =
573 ds->ds_phys->ds_prev_snap_obj;
574 } else if (zap_remove_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
575 ds->ds_object, tx) == 0) {
576 VERIFY(zap_add_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
577 ds->ds_phys->ds_prev_snap_obj, tx) == 0);
582 dsl_pool_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx)
584 dsl_pool_t *dp = ds1->ds_dir->dd_pool;
586 if (dp->dp_scrub_func == SCRUB_FUNC_NONE)
589 if (dp->dp_scrub_bookmark.zb_objset == ds1->ds_object) {
590 dp->dp_scrub_bookmark.zb_objset = ds2->ds_object;
591 } else if (dp->dp_scrub_bookmark.zb_objset == ds2->ds_object) {
592 dp->dp_scrub_bookmark.zb_objset = ds1->ds_object;
595 if (zap_remove_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
596 ds1->ds_object, tx) == 0) {
597 int err = zap_add_int(dp->dp_meta_objset,
598 dp->dp_scrub_queue_obj, ds2->ds_object, tx);
599 VERIFY(err == 0 || err == EEXIST);
601 /* Both were there to begin with */
602 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
603 dp->dp_scrub_queue_obj, ds1->ds_object, tx));
605 } else if (zap_remove_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
606 ds2->ds_object, tx) == 0) {
607 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
608 dp->dp_scrub_queue_obj, ds1->ds_object, tx));
612 struct enqueue_clones_arg {
619 enqueue_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
621 struct enqueue_clones_arg *eca = arg;
626 err = dsl_dataset_hold_obj(spa->spa_dsl_pool, dsobj, FTAG, &ds);
629 dp = ds->ds_dir->dd_pool;
631 if (ds->ds_dir->dd_phys->dd_origin_obj == eca->originobj) {
632 while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) {
634 err = dsl_dataset_hold_obj(dp,
635 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
637 dsl_dataset_rele(ds, FTAG);
642 VERIFY(zap_add_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
643 ds->ds_object, eca->tx) == 0);
645 dsl_dataset_rele(ds, FTAG);
650 scrub_visitds(dsl_pool_t *dp, uint64_t dsobj, dmu_tx_t *tx)
653 uint64_t min_txg_save;
655 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
658 * Iterate over the bps in this ds.
660 min_txg_save = dp->dp_scrub_min_txg;
661 dp->dp_scrub_min_txg =
662 MAX(dp->dp_scrub_min_txg, ds->ds_phys->ds_prev_snap_txg);
663 scrub_visit_rootbp(dp, ds, &ds->ds_phys->ds_bp);
664 dp->dp_scrub_min_txg = min_txg_save;
666 if (dp->dp_scrub_pausing)
670 * Add descendent datasets to work queue.
672 if (ds->ds_phys->ds_next_snap_obj != 0) {
673 VERIFY(zap_add_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
674 ds->ds_phys->ds_next_snap_obj, tx) == 0);
676 if (ds->ds_phys->ds_num_children > 1) {
677 if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
678 struct enqueue_clones_arg eca;
680 eca.originobj = ds->ds_object;
682 (void) dmu_objset_find_spa(ds->ds_dir->dd_pool->dp_spa,
683 NULL, enqueue_clones_cb, &eca, DS_FIND_CHILDREN);
685 VERIFY(zap_join(dp->dp_meta_objset,
686 ds->ds_phys->ds_next_clones_obj,
687 dp->dp_scrub_queue_obj, tx) == 0);
692 dsl_dataset_rele(ds, FTAG);
697 enqueue_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
704 err = dsl_dataset_hold_obj(spa->spa_dsl_pool, dsobj, FTAG, &ds);
708 dp = ds->ds_dir->dd_pool;
710 while (ds->ds_phys->ds_prev_snap_obj != 0) {
712 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
715 dsl_dataset_rele(ds, FTAG);
720 * If this is a clone, we don't need to worry about it for now.
722 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
723 dsl_dataset_rele(ds, FTAG);
724 dsl_dataset_rele(prev, FTAG);
727 dsl_dataset_rele(ds, FTAG);
731 VERIFY(zap_add_int(dp->dp_meta_objset, dp->dp_scrub_queue_obj,
732 ds->ds_object, tx) == 0);
733 dsl_dataset_rele(ds, FTAG);
738 dsl_pool_scrub_sync(dsl_pool_t *dp, dmu_tx_t *tx)
740 spa_t *spa = dp->dp_spa;
743 boolean_t complete = B_TRUE;
745 if (dp->dp_scrub_func == SCRUB_FUNC_NONE)
749 * If the pool is not loaded, or is trying to unload, leave it alone.
751 if (spa->spa_load_state != SPA_LOAD_NONE || spa_shutting_down(spa))
754 if (dp->dp_scrub_restart) {
755 enum scrub_func func = dp->dp_scrub_func;
756 dp->dp_scrub_restart = B_FALSE;
757 dsl_pool_scrub_setup_sync(dp, &func, kcred, tx);
760 if (spa->spa_root_vdev->vdev_stat.vs_scrub_type == 0) {
762 * We must have resumed after rebooting; reset the vdev
763 * stats to know that we're doing a scrub (although it
764 * will think we're just starting now).
766 vdev_scrub_stat_update(spa->spa_root_vdev,
767 dp->dp_scrub_min_txg ? POOL_SCRUB_RESILVER :
768 POOL_SCRUB_EVERYTHING, B_FALSE);
771 dp->dp_scrub_pausing = B_FALSE;
772 dp->dp_scrub_start_time = lbolt64;
773 dp->dp_scrub_isresilver = (dp->dp_scrub_min_txg != 0);
774 spa->spa_scrub_active = B_TRUE;
776 if (dp->dp_scrub_bookmark.zb_objset == 0) {
777 /* First do the MOS & ORIGIN */
778 scrub_visit_rootbp(dp, NULL, &dp->dp_meta_rootbp);
779 if (dp->dp_scrub_pausing)
782 if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
783 VERIFY(0 == dmu_objset_find_spa(spa,
784 NULL, enqueue_cb, tx, DS_FIND_CHILDREN));
786 scrub_visitds(dp, dp->dp_origin_snap->ds_object, tx);
788 ASSERT(!dp->dp_scrub_pausing);
789 } else if (dp->dp_scrub_bookmark.zb_objset != -1ULL) {
791 * If we were paused, continue from here. Note if the
792 * ds we were paused on was deleted, the zb_objset will
793 * be -1, so we will skip this and find a new objset
796 scrub_visitds(dp, dp->dp_scrub_bookmark.zb_objset, tx);
797 if (dp->dp_scrub_pausing)
802 * In case we were paused right at the end of the ds, zero the
803 * bookmark so we don't think that we're still trying to resume.
805 bzero(&dp->dp_scrub_bookmark, sizeof (zbookmark_t));
807 /* keep pulling things out of the zap-object-as-queue */
808 while (zap_cursor_init(&zc, dp->dp_meta_objset, dp->dp_scrub_queue_obj),
809 zap_cursor_retrieve(&zc, &za) == 0) {
810 VERIFY(0 == zap_remove(dp->dp_meta_objset,
811 dp->dp_scrub_queue_obj, za.za_name, tx));
812 scrub_visitds(dp, za.za_first_integer, tx);
813 if (dp->dp_scrub_pausing)
815 zap_cursor_fini(&zc);
817 zap_cursor_fini(&zc);
818 if (dp->dp_scrub_pausing)
823 dsl_pool_scrub_cancel_sync(dp, &complete, kcred, tx);
826 VERIFY(0 == zap_update(dp->dp_meta_objset,
827 DMU_POOL_DIRECTORY_OBJECT,
828 DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
829 &dp->dp_scrub_bookmark, tx));
830 VERIFY(0 == zap_update(dp->dp_meta_objset,
831 DMU_POOL_DIRECTORY_OBJECT,
832 DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
833 &spa->spa_scrub_errors, tx));
835 /* XXX this is scrub-clean specific */
836 mutex_enter(&spa->spa_scrub_lock);
837 while (spa->spa_scrub_inflight > 0)
838 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
839 mutex_exit(&spa->spa_scrub_lock);
843 dsl_pool_scrub_restart(dsl_pool_t *dp)
845 mutex_enter(&dp->dp_scrub_cancel_lock);
846 dp->dp_scrub_restart = B_TRUE;
847 mutex_exit(&dp->dp_scrub_cancel_lock);
855 count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp)
860 * If we resume after a reboot, zab will be NULL; don't record
861 * incomplete stats in that case.
866 for (i = 0; i < 4; i++) {
867 int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS;
868 int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL;
869 zfs_blkstat_t *zb = &zab->zab_type[l][t];
873 zb->zb_asize += BP_GET_ASIZE(bp);
874 zb->zb_lsize += BP_GET_LSIZE(bp);
875 zb->zb_psize += BP_GET_PSIZE(bp);
876 zb->zb_gangs += BP_COUNT_GANG(bp);
878 switch (BP_GET_NDVAS(bp)) {
880 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
881 DVA_GET_VDEV(&bp->blk_dva[1]))
882 zb->zb_ditto_2_of_2_samevdev++;
885 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
886 DVA_GET_VDEV(&bp->blk_dva[1])) +
887 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
888 DVA_GET_VDEV(&bp->blk_dva[2])) +
889 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
890 DVA_GET_VDEV(&bp->blk_dva[2]));
892 zb->zb_ditto_2_of_3_samevdev++;
894 zb->zb_ditto_3_of_3_samevdev++;
901 dsl_pool_scrub_clean_done(zio_t *zio)
903 spa_t *spa = zio->io_spa;
905 zio_data_buf_free(zio->io_data, zio->io_size);
907 mutex_enter(&spa->spa_scrub_lock);
908 spa->spa_scrub_inflight--;
909 cv_broadcast(&spa->spa_scrub_io_cv);
911 if (zio->io_error && (zio->io_error != ECKSUM ||
912 !(zio->io_flags & ZIO_FLAG_SPECULATIVE)))
913 spa->spa_scrub_errors++;
914 mutex_exit(&spa->spa_scrub_lock);
918 dsl_pool_scrub_clean_cb(dsl_pool_t *dp,
919 const blkptr_t *bp, const zbookmark_t *zb)
921 size_t size = BP_GET_PSIZE(bp);
922 spa_t *spa = dp->dp_spa;
924 int zio_flags = ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
927 ASSERT(bp->blk_birth > dp->dp_scrub_min_txg);
929 if (bp->blk_birth >= dp->dp_scrub_max_txg)
932 count_block(dp->dp_blkstats, bp);
934 if (dp->dp_scrub_isresilver == 0) {
936 zio_flags |= ZIO_FLAG_SCRUB;
937 zio_priority = ZIO_PRIORITY_SCRUB;
940 /* It's a resilver */
941 zio_flags |= ZIO_FLAG_RESILVER;
942 zio_priority = ZIO_PRIORITY_RESILVER;
946 /* If it's an intent log block, failure is expected. */
947 if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
948 zio_flags |= ZIO_FLAG_SPECULATIVE;
950 for (int d = 0; d < BP_GET_NDVAS(bp); d++) {
951 vdev_t *vd = vdev_lookup_top(spa,
952 DVA_GET_VDEV(&bp->blk_dva[d]));
955 * Keep track of how much data we've examined so that
956 * zpool(1M) status can make useful progress reports.
958 mutex_enter(&vd->vdev_stat_lock);
959 vd->vdev_stat.vs_scrub_examined +=
960 DVA_GET_ASIZE(&bp->blk_dva[d]);
961 mutex_exit(&vd->vdev_stat_lock);
963 /* if it's a resilver, this may not be in the target range */
965 if (DVA_GET_GANG(&bp->blk_dva[d])) {
967 * Gang members may be spread across multiple
968 * vdevs, so the best estimate we have is the
969 * scrub range, which has already been checked.
970 * XXX -- it would be better to change our
971 * allocation policy to ensure that all
972 * gang members reside on the same vdev.
976 needs_io = vdev_dtl_contains(vd, DTL_PARTIAL,
982 if (needs_io && !zfs_no_scrub_io) {
983 void *data = zio_data_buf_alloc(size);
985 mutex_enter(&spa->spa_scrub_lock);
986 while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight)
987 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
988 spa->spa_scrub_inflight++;
989 mutex_exit(&spa->spa_scrub_lock);
991 zio_nowait(zio_read(NULL, spa, bp, data, size,
992 dsl_pool_scrub_clean_done, NULL, zio_priority,
996 /* do not relocate this block */
1001 dsl_pool_scrub_clean(dsl_pool_t *dp)
1004 * Purge all vdev caches. We do this here rather than in sync
1005 * context because this requires a writer lock on the spa_config
1006 * lock, which we can't do from sync context. The
1007 * spa_scrub_reopen flag indicates that vdev_open() should not
1008 * attempt to start another scrub.
1010 spa_config_enter(dp->dp_spa, SCL_ALL, FTAG, RW_WRITER);
1011 dp->dp_spa->spa_scrub_reopen = B_TRUE;
1012 vdev_reopen(dp->dp_spa->spa_root_vdev);
1013 dp->dp_spa->spa_scrub_reopen = B_FALSE;
1014 spa_config_exit(dp->dp_spa, SCL_ALL, FTAG);
1016 return (dsl_pool_scrub_setup(dp, SCRUB_FUNC_CLEAN));